Cographs; chordal graphs and tree

decompositions
Zdeněk Dvořák
September 14, 2015

Let us now proceed with some more interesting graph classes closed on
induced subgraphs.

1 Cographs
The class of cographs is defined recursively as follows.

• The graph with one vertex is a cograph.

• The disjoint union of two cographs is a cograph.

• The complement of a cograph is a cograph.

A graph is a cograph if and only if it can be obtained by a finite number

of applications of these rules. For example, K4 is a cograph, because its
complement is a union of single-vertex graphs.

Theorem 1. Forbv (P4 ) = cographs.

Proof. Firstly, note that the class of cographs is induced-subgraph-closed.

Furthermore, P4 is not a cograph, since it has more than one vertex and both
P4 and its complement are connected. Hence, if a graph G contains P4 as an
induced subgraph, then G is not a cograph, and thus Forbv (P4 ) ⊆ cographs.
Therefore, we only need to prove that every graph G such that P4 6v G
is a cograph. We prove the claim by induction, and thus we assume that the
claim holds for all graphs with less than |V (G)| vertices. Since all graphs
with at most 3 vertices are cographs, we can assume that |V (G)| ≥ 4. If
G is not connected, then by induction hypothesis, all the components of G
are cographs, and thus G is a cograph. Hence, assume that G is connected.

1
Consider any vertex v ∈ V (G), and let G0 = G − v. Note that P4 6v G0 , and
by the induction hypothesis, G0 is a cograph.
Suppose first that G0 is not connected, and let C1 , . . . , Cn be the compo-
nents of G0 . Since G is connected, v has a neighbor ui ∈ V (Ci ) for 1 ≤ i ≤ n.
If v is adjacent to all vertices of G0 , then G = G0 ∪ {v} is a cograph. Hence,
we can assume that a vertex w1 ∈ V (C1 ) is not adjacent to v. Since C1 is
connected, we can assume that u1 is adjacent to w1 . However, then w1 u1 vv2
is an induced path in G, which contradicts the assumption that P4 6v G.
Finally, consider the case that G0 is connected. Since G0 is a cograph
with more than one vertex, its complement is not connected. Furthermore,
P4 = P4 , and thus P4 6v G. By the same argument as in the previous
paragraph, we prove that G is a cograph, and thus G is a cograph.
A graph G is chordal if no cycle of length greater than 3 is induced,
that is, chordal = Forbv (holes). There are several related characterization
of chordal graphs.
For a connected graph G, S ⊆ V (G) is a minimal cut if G − S is not
connected, but G − S 0 is connected for every S 0 ( S. Observe that every
vertex of a minimal cut S has a neighbor in every component of G − S.
Lemma 2. If a connected graph G is chordal, then every minimal cut S ⊆
V (G) induces a clique.
Proof. Suppose that u, v ∈ S are not adjacent. Let C1 and C2 be components
of G − S. Since S is a minimal cut, both u and v have neighbors both in C1
and in C2 . For i ∈ {1, 2}, let Pi be a shortest path between u and v through
Ci . Then P1 ∪ P2 is an induced hole in G, which contradicts the assumption
that G is chordal.
Theorem 3. If G is chordal and not a clique, then there exist G1 , G2 @ G
such that G = G1 ∪ G2 and G1 ∩ G2 is a clique.
Proof. If G is not connected, we can take G1 to be a component of G and
G2 = G \ V (G2 ). Hence, assume that G is connected. Since G is not a clique,
there exist vertices x, y ∈ V (G) that are not adjacent. Let S0 = V (G)\{x, y}.
Then G − S0 is not connected. Consequently, there exists S ⊆ S0 such that
S is a minimal cut. Let G − S = C1 ∪ C2 , where C1 and C2 are non-empty
and disjoint. Then we can set G1 = G − V (C2 ) and G2 = G − V (C1 ). Note
that G1 ∩ G2 = G[S] is a clique by Lemma 2.
Observe also that if G1 and G2 are chordal and G1 ∩ G2 is a clique, then
G1 ∪ G2 is chordal. Hence, every chordal graph can be obtained using a finite
number of applications of these rules:

2
 • A complete graph is chordal.

• If G1 and G2 are chordal and G is obtained from G1 and G2 by gluing

on a clique, then G is chordal.

A vertex v ∈ V (G) is simplicial if the neighborhood of v in G induces a

clique.

Lemma 4. If G is chordal and not a clique, then G contains two non-

adjacent simplicial vertices.

Proof. We prove the claim by induction, and thus we assume that it holds
for all graphs with less than |V (G)| vertices. By Theorem 3, there exist
G1 , G2 @ G such that G = G1 ∪ G2 and G1 ∩ G2 is a clique. For i ∈ {1, 2},
if Gi is not a clique, then by the induction hypothesis, Gi contains two non-
adjacent simplicial vertices, and at most one of them belongs to the clique
G1 ∩ G2 . Let vi be a simplicial vertex of Gi not belonging to G1 ∩ G2 . If Gi
is a clique, then let vi be any vertex of Gi not belonging to G1 ∩ G2 .
Since neither v1 nor v2 belong to G1 ∩ G2 , observe that v1 and v2 are
non-adjacent simplicial vertices of G.

Theorem 5. A graph G is chordal if and only if every induced subgraph of

G contains a simplicial vertex.

Proof. Every induced subgraph of a chordal graph is chordal, and contains

a simplicial vertex by Lemma 4. Hence, it suffices to prove that if every
induced subgraph of G contains a simplicial vertex, then G is chordal. We
prove the claim by induction on the number of vertices of G, and thus we
assume that the claim holds for all graphs with less than |V (G)| vertices.
Let v be a simplicial vertex of G. By the induction hypothesis, G − v is
chordal, and thus every induced hole in G contains v. However, for any hole
C = uvwz1 z2 . . . ⊆ G, the vertices u and w are adjacent, and thus C is not
induced. It follows that G is chordal.
A tree decomposition of a graph G is a pair (T, β), where β : V (T ) → 2V (G)
assigns a bag β(n) to each vertex of T , such that

• for every v ∈ V (G), there exists n ∈ V (T ) such that v ∈ β(n) – “every

vertex is in some bag”,

• for every uv ∈ E(G), there exists n ∈ V (T ) such that u, v ∈ β(n) –

“every edge is in some bag”, and

3
 • for every v ∈ V (G), the set {n ∈ V (T ) : v ∈ β(n)} induces a connected
subtree of T – “every vertex appears in a connected subtree of the
decomposition”.

The width of the tree decomposition is the size of the largest bag minus
one. Example: a tree decomposition of C12 of width 4.
v3
v4 v2
v1 v2 v3 v4 v5 v6
v5 v1
v1 v3 v4 v6 v12
v6 v12
v6 v7 v9 v10 v12
v7 v11
v7 v8 v9 v10 v11 v12
v8 v10
v9

A tree decomposition (T, β) is reduced if every n1 n2 ∈ E(T ) satisfies

β(n1 ) 6⊆ β(n2 ) and β(n2 ) 6⊆ β(n1 ).

Lemma 6. For every tree decomposition (T, β) of G, there exists a reduced

tree decomposition (T 0 , β 0 ) of G such that each bag of (T 0 , β 0 ) is equal to some
bag of (T, β).

Proof. Suppose that n1 n2 ∈ E(T ) satisfies β(n1 ) ⊆ β(n2 ). Let T1 be the tree
obtained from T by contracting the edge n1 n2 , and let n ∈ V (T1 ) be the
vertex obtained from n1 and n2 by this contraction. Let β1 : V (T1 ) → 2V (G)
be defined by β1 (n) = β(n2 ) and β1 (n0 ) = β(n0 ) for every n0 ∈ V (T1 ) \
{n}. Then (T1 , β1 ) is a tree decomposition of G with fewer vertices. By
repeating this operation as long as possible, we eventually obtain a reduced
tree decomposition of G.

Corollary 7. If G has a tree decomposition of width at most k, then the

minimum degree of G is at most k.

Proof. Let (T, β) be a reduced tree decomposition of G of width at most k,

and let n be a leaf of T . Since (T, β) is reduced, there exists v ∈ V (G) such
that v only appears in the bag of n. Hence, all neighbors of v belong to β(n),
and thus deg(v) ≤ |β(n)| − 1 ≤ k.

Lemma 8. If G is a complete graph, then G has only one reduced tree de-
composition: tree with one vertex with bag equal to V (G).

4
Proof. Let (T, β) be a reduced tree decomposition of G, and suppose that T
has an edge n1 n2 . Since (T, β) is reduced, there exists x ∈ β(n1 ) \ β(n2 ) and
y ∈ β(n2 ) \ β(n1 ). Let Tx be the subtree of T induced by the vertices whose
bags contain x, and let Ty be the subtree of T induced by the vertices whose
bags contain y. Since n1 6∈ V (Tx ) and n2 6∈ V (Ty ), the edge n1 n2 belongs
neither to Tx nor to Ty . Hence, Tx and Ty are subgraphs of different compo-
nents of T − n1 n2 , and thus they are disjoint. However, xy ∈ E(G) implies
that some bag contains both x and y, which is a contradiction. Therefore, a
reduced tree decomposition of G has no edges.

Corollary 9. If S ⊆ V (G) induces a clique in G and (T, β) is a tree decom-

position of G, then there exists n ∈ V (T ) such that S ⊆ β(n).

Proof. Let β1 : V (T ) → 2S be defined by β1 (n) = β(n) ∩ S for every n ∈

V (T ). Then (T, β1 ) is a tree decomposition of the clique G[S]. By Lemma 6,
there exists a reduced tree decomposition (T2 , β2 ) of G[S] whose every bag
is equal to some bag of (T, β1 ). By Lemma 8, T2 has only one vertex n2 ,
and β2 (n2 ) = S. Let n be a vertex of T such that β2 (n2 ) = β1 (n). We have
S = β2 (n2 ) = β1 (n) ⊆ β(n).

Theorem 10. A graph G is chordal if and only if it has a tree decomposition

(T, β) such that every bag induces a clique.

Proof. Suppose that G is chordal. We proceed by induction, and thus we

can assume that the claim holds for all graphs with less than |V (G)| vertices.
If G is a clique, then we can set V (T ) = {n} and β(n) = V (G). Hence,
assume that G is not a clique, and by Theorem 3, there exist G1 , G2 @ G
such that G = G1 ∪ G2 and G1 ∩ G2 is a clique. For i ∈ {1, 2}, the induction
hypothesis implies that there exists a tree decomposition (Ti , βi ) of Gi such
that every bag induces a clique. By Corollary 9, there exists ni ∈ V (Ti ) such
that V (G1 ∩ G2 ) ⊆ βi (ni ). Let T be the tree obtained from T1 and T2 by
adding the edge n1 n2 . Let β be defined by β(n) = β1 (n) if n ∈ V (T1 ) and
β(n) = β2 (n) if n ∈ V (T2 ). Then (T, β) is a tree decomposition of G such
that every bag induces a clique.
Conversely, suppose that G has a tree decomposition (T, β) such that
every bag induces a clique. Consider a cycle C v G, and let β 0 : V (T ) →
2V (C) be defined by β 0 (n) = β(n)∩V (C) for every n ∈ V (T ). Then, (T, β 0 ) is a
tree decomposition of C such that every bag induces a clique. By Corollary 7,
any tree decomposition of a cycle must contain a bag of size at least three,
and thus C contains a clique of size three. Therefore, C is a triangle. It
follows that G contains no induced hole, and thus G is chordal.

5
2 Exercises
1. (?) Let C be the class of graphs that can be obtained by a finite number
of applications of these rules:

• The graph with one vertex belongs to C.

• For any G1 , G2 ∈ C, the disjoint union of G1 and G2 belongs to C.
• For any G1 , G2 ∈ C, the complete join of G1 and G2 (the graph
obtained from the disjoint union of G1 and G2 by adding all edges
with one end in G1 and the other end in G2 ) belongs to C.

Prove that C is exactly the class of all cographs.

2. (? ? ?) A graph G is a split graph if its vertex set can be partitioned

to an independent set and a clique (with arbitrary edges between the
two parts), i.e., there exist disjoint A, B ⊆ V (G) such that A ∪ B =
V (G), G[A] is an independent set and G[B] is a clique. Prove that
Forbv (C4 , C5 , 2K2 ) = split graphs. Hint: first show that all graphs in
Forbv (C4 , C5 , 2K2 ) are chordal.

3. (??) Let (T, β) be a tree decomposition of G. Let n1 , n2 and n3 be

vertices of T such that n2 lies on the path between n1 and n3 in T .
Prove that every path in G from β(n1 ) to β(n3 ) intersects β(n2 ).